
% Author: Kazuma Takakura
% Date: 17 May 2025
% Time: 10:51:35

\begin{table}[h!]\footnotesize
  \centering
  \caption{Heterogeneity among Baseline Abilites (Matched Sample)}
\label{tab:hetero_match}
\scalebox{1}{
\begin{threeparttable}
\begin{tabular}{lccc}\toprule
  & Top 50\%^{b}  & Bottom 50\%^{b} & Differences  \\\midrule
\multicolumn{4}{c}{Panel A: Rapid math test score^{a}}\\\midrule
 Treatment & -0.476**  & -0.200 &  -0.276 \\
 & ( 0.036 ) & ( 0.490 ) & ( 0.382 ) \\ 
 Observation &  95 & 95 &  \\ 
 N of clusters &  27 & 27 &  \\\midrule 
\multicolumn{4}{c}{Panel B: RSES score^{a}}\\\midrule
 Treatment & 1.116***  & 0.205 &  0.911 *** \\
 & ( 0.000 ) & ( 0.494 ) & ( 0.006 ) \\ 
 Observation &  82 & 102 &  \\ 
 N of clusters &  25 & 27 &  \\\midrule 
\multicolumn{4}{c}{Panel C: CPCS score^{a}}\\\midrule
 Treatment & 0.863***  & 0.476 &  0.387 \\
 & ( 0.006 ) & ( 0.144 ) & ( 0.284 ) \\ 
 Observation &  71 & 113 &  \\ 
 N of clusters &  26 & 26 &  \\\midrule 
\end{tabular}
\begin{tablenotes}
\item (a) Dependent variables are standardized using the average and variance of the whole follow-up sample in the February 2022 survey.
\item (b) Cutoffs are created based on whether their ability to perform each item at the baseline was higher or lower than the median. 
\item (c) Wild cluster bootstrap p-values are reported within parentheses. Clusters are schools at the baseline. 
\item (d) $^*$ Significant at 10\% level; $^{**}$ significant at 5\% level; $^{***}$ significant at 1\% level. 
\end{tablenotes}
\end{threeparttable}
}
\label{tab:addlabel}%
\end{table}




